Input, Tidy, Model, Output

Use RStudio projects!

• scripts/
• data/

Scripts

In scripts/ folder:

• scripts/input.R - download and gather data
• scripts/tidy.R - process data into usable by model step
• scripts/model.R - create models
• scripts/out.R - make plots, Rmd, final outputs

Data

In data/ folder:

• data/raw/ - any raw data files (CSVs etc)
• data/cache/ - API cache files
• data/output/ - clean tidy files

Use saveRDS/readRDS rather than write.csv()/read.csv() - 10xquicker

Why ga_model()?

• Advanced R users sharing results quickly with newbie R users
• Reproduceable objects
• Enables quick Shiny deployments
• Tweet embeds

ga_model example

library(prophet)
library(googleAnalyticsR)
library(googleAnalyticsModelR)

ga_auth(email = Sys.getenv("GARGLE_EMAIL"))
my_view_id <- 81416156

forecast_data <- ga_model_prophet(my_view_id, interactive_plot = TRUE)
forecast_data$plot

ga_model example output

Decomposition example

my_viewid <- 81416156

# load the model (equivalent to ga_model_load())
decomp_ga <- ga_model_example("decomp_ga.gamr")

# apply model to your data
d1 <- ga_model(my_viewid, model = decomp_ga)

Decomposition example output

ga_model object structure

• decomp_ga$data_f = function to get data
• decomp_ga$model_f = function to model data
• decomp_ga$output_f = function to plot data

decomp_ga$model_f

## function(df, ...){
##     decompose(ts(df$sessions, frequency = 7))
## }

Build your own ga_model - step 1

• Create a data fetching function

fetch_data <- function(viewId, ...){
    google_analytics(viewId, 
                     date_range = c("10DaysAgo", "yesterday"),
                     metrics = "sessions", 
                     dimensions = "date",
                     max = -1)
}

Build your own ga_model - step 2

• Create a modelling function

model_data <- function(df, ...){
    decompose(ts(df$sessions, frequency=7))
}

Build your own ga_model - step 3

• Create an output function

plot_data <- function(df, ...){
    plot(df)
}

Build your own ga_model - step 4

• Build model object with ga_model_make()

decomp_ga <- ga_model_make(data_f = fetch_data,
                           model_f = model_data,
                           output_f = plot_data,
                           required_columns = c("date", "sessions"),
                           description = "Performs decomposition and creates a plot")

Build your own ga_model - step 5 (optional)

• Build model object with Shiny outputs

decomp_ga <- ga_model_make(get_model_data,
                           model_f = decompose_sessions,
                           output_f = graphics::plot,
                           required_columns = c("date", "sessions"),
                           description = "Performs decomposition and creates a plot",
                           outputShiny = shiny::plotOutput,
                           renderShiny = shiny::renderPlot)

Build your own ga_model - step 6

• Save model object to a .gamr object

ga_model_save(decomp_ga, "my-decomp-model.gamr")

Use your ga_model

my_model <- ga_model_load("my-decomp-model.gamr")
my_viewid <- 81234124
ga_model(my_viewid, my_model)

YouTube video!

A bit out of date!

Creating Shiny deployments - auth

library(shiny)
library(googleAuthR)
library(googleAnalyticsR)
                             
ui <- fluidPage(
    titlePanel("ga_model Shiny Demo"),
    authDropdownUI("picker"),
)

server <- function(input, output, session) {

    gar_shiny_auth(session)
    viewId <- callModule(authDropdown, "picker",
                         ga.table = reactive(ga_account_list()))

}

# Run the application
shinyApp(ui = gar_shiny_ui(ui), server = server)

Creating Shiny deployments - ga_model

library(shiny)
library(googleAnalyticsR)
library(googleAuthR)

decomp_ga <- ga_model_example("decomp_ga.gamr")
shiny_model <- ga_model_edit(decomp_ga,
                             outputShiny = plotOutput,
                             renderShiny = renderPlot)
                             
ui <- fluidPage(
    titlePanel("ga_model Shiny Demo"),
    authDropdownUI("picker"),
    shiny_model$shiny_module$ui("model1")
)

server <- function(input, output, session) {

    gar_shiny_auth(session)
    viewId <- callModule(authDropdown, "picker",
                         ga.table = reactive(ga_account_list()))
    callModule(shiny_model$shiny_module$server, "model1",
               view_id = viewId)

}

# Run the application
shinyApp(ui = gar_shiny_ui(ui), server = server)

Recommended libraries for using with googleAnalyticsR

• forecast
• prophet
• CausalImpact
• funneljoin